LinkedIn MCP Server: Give Your AI Agent LinkedIn Access
Point Claude, ChatGPT, or any AI agent at LinkedIn: find people, send invites and messages, and pull enriched profiles, on accounts that stay safe by design.
A LinkedIn MCP server is a typed set of tools an AI agent calls to act on LinkedIn: search, connect, message, and enrich. GTM API is one, built for agents and run on your own accounts: connect Claude, ChatGPT, Cursor or any other MCP client to one endpoint and your agent runs real LinkedIn outreach through 190+ typed tools, with account safety enforced server-side. It’s the same infrastructure that runs 20,000+ LinkedIn accounts, with under 1% of them ever restricted.
Drafting the message is the easy half. An MCP server is the layer that lets the agent do the work itself (send the invite, reply in the thread, enrich the lead) with no browser extension and no cookie scraping. The rest of this page covers what an MCP server is, how to connect it to Claude, what the agent can do once it is connected, and where to start in the docs.
On this page
What is a LinkedIn MCP server?
A LinkedIn MCP server is a program that exposes LinkedIn actions (search, connect, message, enrich) as Model Context Protocol tools that an AI agent can call. Instead of a person clicking through a dashboard, the agent reads the tool list and runs the action itself. A managed one is GTM API: hosted and typed, wired into the safety stack that keeps accounts alive.
MCP is the open standard introduced by Anthropic in late 2024 and since adopted across Claude, OpenAI’s ChatGPT, Cursor, and the wider agent ecosystem. People describe it as “a USB-C port for AI”: one connector that lets any model plug into any tool. A LinkedIn MCP server applies that standard to a single channel, LinkedIn.
At its core, GTM API is MCP-first: every capability is defined once as a typed schema, which generates the MCP tools, a REST API, and webhooks from the same contract. An agent in any MCP client (Claude, ChatGPT, Cursor, VS Code and the rest) and a custom backend over REST therefore act on LinkedIn through exactly the same validated operations, with no drift between them and no glue code to maintain.
How to connect LinkedIn to Claude
Connect LinkedIn to Claude in two steps: open Settings → Connectors → “Add custom connector”, paste https://mcp.gtm-api.com/mcp, and approve the consent screen that opens in your browser. There is no key to paste. The server advertises its authorization server and your client registers itself, so consent is the whole credential. Claude Code takes one command, claude mcp add --transport http gtm-api https://mcp.gtm-api.com/mcp, then /mcp in a session to authenticate. Per-client setup, the config-file trap and troubleshooting are on Connect LinkedIn to Claude. Any other MCP client takes the same URL, and the docs page Connect a client has the steps for ChatGPT, Cursor, VS Code and Codex.
Prefer a local CLI?
Clients that only speak stdio, and headless agents with no browser for the consent screen, launch the same server locally through the npm CLI. Create an API key in the app, then it is one line, no clone, no config file:
GTM_API_KEY=gtm_live_xxx npx -y @gtm-api/linkedin-mcp
Then prompt in plain English
Once the connector reports connected, just ask. The agent sees the LinkedIn tools and picks the right ones:
“Every morning, accept new connection invitations from founders, reply with a short welcome, and add anyone hiring SDRs to a ‘warm’ list.”
Under the hood the agent calls get_my_latest_linkedin_connection_invitations, accept_linkedin_connection_invitation, and send_linkedin_message. Every outward action runs a preview → confirm step and a server-side daily-limit check before it ever touches LinkedIn.
What can the agent do?
A GTM API agent can run the full LinkedIn outreach loop: find people, send and accept connections, message across every LinkedIn surface, publish and engage with posts, enrich profiles and companies, and read its own account health. That’s 190+ typed tools across 12 LinkedIn toolsets, all behind one key.
Sending is real: send_linkedin_message, send_linkedin_connection_request and send_linkedin_inmail go out from your own account, the agent reads the reply with get_my_latest_linkedin_messages, and marking a thread unread hands it back to a person when judgment is needed. The same actions over REST, and how other automation APIs price and pace them, are on LinkedIn automation API.
| Toolset | What your agent can do |
|---|---|
| Messaging | Send DMs, voice notes, InMail and Sales Navigator messages. Sync and search the whole inbox. Read reply metrics |
| Network | Send & withdraw connection requests, accept/ignore invitations, list connections, followers and their growth |
| Content | Publish posts, comment, react, and endorse a connection’s skills |
| Recruiter | Read a Recruiter seat’s inbox and hiring projects, send Recruiter InMails |
| Scraping / search | People, company and post search by URL or params. Similar profiles, employees, decision-makers, post engagers |
| Auto-scrapes | Saved searches that rerun on a schedule, dedupe across runs and hand new leads to a mass action |
| Enrichment | Lite & full profile, experience, skills, education, posts, plus company data, returned as typed JSON |
| Data | The data-request ledger: every scrape and enrichment call as one row, to audit and re-read |
| Accounts & pacing | Connect accounts, set smart daily limits, read the self-profile, reset a stale sync |
| Account monitor | Health snapshots, benchmarks, quota-hit log, block log, activity log, sync runs (read-only audit) |
| Browsers | Isolated cloud browsers and dedicated proxies with connectivity & location checks |
| Platform | An admin-gated custom request for LinkedIn endpoints the typed tools do not cover |
Outside the LinkedIn toolsets, the same server carries webhooks with their delivery log, and mass actions for running one step across many targets under one approval.
LinkedIn is the live channel today. Email (Gmail, Outlook, IMAP), messengers (WhatsApp, Telegram, Instagram DMs) and calendars (Google, Microsoft) are on the roadmap on the same typed contract, so an agent built against LinkedIn today will pick up those channels as they land.
What is a LinkedIn AI agent? covers what an agent cannot do on LinkedIn and how unattended it can really run.
Where to start: the AI Assistant in the docs
If you know the outcome but not the tools, start in the docs. The AI Assistant there is wired to the live API and the changelog: describe the play in plain English, like “find posts about a keyword and pull everyone who commented”, and it answers with the endpoints and the order to call them in, each linked to its reference page. It works from the docs alone and needs no account. The walkthrough with screenshots is on The gtm-api.com AI Assistant.
That is the whole setup: one URL in the client, a consent screen, and the agent works on LinkedIn through typed tools, with the account’s limits checked on the server before every call. To try it, one LinkedIn account runs on the free plan, with the same tools a paid account gets and no card.
Frequently Asked Questions
Is there a LinkedIn MCP server?
Yes: GTM API is a managed LinkedIn MCP server, a hosted set of Model Context Protocol tools an AI agent calls to search, connect, message and enrich on LinkedIn. There are also open-source LinkedIn MCP servers on GitHub, but they scrape with your cookie and have no safety layer, so they are better for demos than production.
Which AI clients support the GTM API LinkedIn MCP?
Claude (claude.ai, Desktop and Claude Code), ChatGPT, Cursor, VS Code, Codex, and any other MCP-compatible client, LangChain and n8n included. For everything else, the same actions are available over REST and webhooks, because GTM API is built from one typed schema that generates MCP, REST and webhooks together.
Do I need to write code to use it?
No. In an MCP client like Claude or Cursor you connect the server once, then prompt in plain English. The agent discovers the LinkedIn tools and runs them. Developers who want to build a product on top can call the same actions over the REST API instead.
Can an AI agent actually send LinkedIn messages?
Yes. Through the MCP server the agent calls real tools, send_linkedin_message, send_linkedin_connection_request, send_linkedin_inmail, so the message is sent from your connected account, not just drafted. Every send runs a preview-then-confirm step and a server-side daily-limit check first.
Is a LinkedIn MCP server safe for my account?
No server can make it risk-free: LinkedIn’s User Agreement does not allow third-party automation, so a plain yes would be wrong. What a server controls is how much risk it adds. GTM API runs each account in its own isolated cloud browser with a dedicated residential proxy, warms new accounts up, and checks health-aware daily limits before any action. On that stack, gtm-api.com runs 20,000+ LinkedIn accounts with under 1% of them ever restricted. Cookie-based open-source servers offer none of that. The full method is on the safe LinkedIn automation page.
Can I connect more than one LinkedIn account?
Yes, that is the design. Connect as many accounts as you like. Each runs in an isolated cloud browser with its own proxy and its own pacing. You pay per connected account per month, and your agent can act across all of them through one key.
How much does the GTM API LinkedIn MCP server cost?
From $10 per connected account per month at volume, with unlimited API calls and no per-action fees. The MCP server, REST and webhooks come with the account, so the number of messages your agent sends does not change the price. Scraping and enrichment are live, with pricing not yet published. The full breakdown is on the LinkedIn API pricing page.
Is there a free trial?
There’s no trial because the free plan doesn’t expire: Sandbox, the forever free plan, covers one connected account with full API and MCP access, no credit card required and no minimum commitment. Pricing is from $10 per connected account per month at volume, with unlimited API calls.
Is LinkedIn automation against LinkedIn’s terms?
Yes. LinkedIn’s User Agreement does not allow bots or other automated access, so any tool that acts on an account carries that risk, and no vendor can honestly promise otherwise. How much risk a given server adds is the safety question above.
Read next on this topic
The gtm-api.com AI AssistantDescribe the play in plain English and let it name the endpoints.Read →
The gtm-api Free PlanOne LinkedIn account, the full UI, API and MCP, no card.Read →
What Is a LinkedIn AI Agent?Model quality writes the message. The account survives on everything else.Read →
Sources & Further Reading
- Anthropic / Model Context Protocol, modelcontextprotocol.io (the open standard)
- gtm-api.com, product & live LinkedIn toolsets
- Example open-source server, stickerdaniel/linkedin-mcp-server (GitHub) · directories: mcpservers.org, MCP Market · our listings: Glama, Smithery
- Related: The LinkedIn API · Safe LinkedIn automation · LinkedIn API pricing · LinkedIn API access · Connect LinkedIn to Claude · LinkedIn integrations
Give your agent LinkedIn access.
Sign up, connect your first LinkedIn account, and point Claude, ChatGPT, Cursor or any MCP client at it in minutes. 190+ typed tools, one contract, the same stack behind the published restriction record.
Last updated: October 2026 · Definitions and benchmarks verified at time of publication